Vehicle path planning interaction method based on vehicle-road cloud collaboration

Abstract

The invention discloses a vehicle path planning interaction method based on vehicle-road cloud collaboration, and the main conception is that a communication link is decomposed into multiple paths of signals, the signals are dynamically switched according to real-time signal quality, seamless degradation is realized, and based on a three-stage elastic architecture of cloud global planning, marginal layer region collaboration and vehicle end local execution, the vehicle path planning interaction method based on the marginal layer region collaboration is realized. A communication degradation control and computing power resource dynamic allocation strategy is utilized to avoid contradiction generated by communication delay and computing power allocation; besides, based on a vehicle-road cloud collaborative architecture and in combination with a space-time consistency guarantee scheme, on the premise of ensuring that the vehicle end can autonomously change a lane during communication interruption, the safety problems that the deviation between the vehicle end autonomous decision and a global path is too large and rapid correction is difficult after communication recovery are solved. According to the invention, aiming at the specific working condition that the communication link is unstable and the computing power resource is limited, the cost benefit of the automatic driving algorithm is obviously optimized, and the driving safety is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle-road cooperation automatic driving, and particularly relates to a vehicle path planning interaction method based on vehicle-road cloud cooperation. BACKGROUND

[0002] In the path planning scene, when the automatic driving vehicle makes a cloud decision, the vehicle lane changing is interrupted under network fluctuation. How to make a vehicle-road cloud cooperation lane changing for the automatic driving vehicle.

[0003] In the current vehicle-road cloud cooperation scheme in the industry, the vehicle-road cloud cooperation subsystem mainly includes a vehicle end, a roadside and a cloud end. The vehicle end is used to obtain vehicle end perception information and transmit the information to the cloud end. The roadside is used to obtain roadside perception information and transmit the information to the cloud end. The cloud end is used to receive the vehicle end and roadside perception information, perform data fusion and storage, and perform real-time data transmission with a travel service subsystem through an interface calling mode. Meanwhile, according to the transmission data of the travel service subsystem, vehicle screening and route planning are performed to make vehicle cooperation decision and remote control.

[0004] As can be seen, the vehicle end and the roadside in the existing system are only used to report perception data, and the data reception, data fusion and analysis are completed by the cloud end to realize vehicle intelligent screening, path planning and decision control in the travel service subsystem. In a low traffic density scene, a large amount of cloud computing power is occupied, and the cost benefit is low. Moreover, the scheme only relies on the cloud end for decision control, and completely relies on a single link network. As described above, in an unstable network area, the cloud end instruction cannot be timely conveyed, resulting in lane changing interruption and even accidents. In addition, in the passing fault and recovery process, the sudden change of the lane changing strategy is also easy to cause the vehicle to lose control. SUMMARY

[0005] In view of the above, the present application aims to provide a vehicle path planning interaction method based on vehicle-road cloud cooperation to solve the aforementioned technical problems.

[0006] The technical solution adopted by the present application is as follows:

[0007] The present application provides a vehicle path planning interaction method based on vehicle-road cloud cooperation, which comprises the following steps:

[0008] In the automatic driving process, if it is detected that the communication link quality is deteriorated, the communication mode is dynamically switched according to the preset communication priority: the current communication mode is downgraded to other interaction mode;

[0009] Based on the communication quality and / or computing power resource state after the downgrade, the path planning decision level related to vehicle lane changing is dynamically allocated;

[0010] According to the assigned path planning decision level, it is determined whether to execute a vehicle-end autonomous mode, which refers to autonomously planning a driving path and a lane-changing strategy by the vehicle end;

[0011] If the vehicle-end autonomous mode has been executed, the trajectory autonomously generated by the vehicle end is checked with the global path through a space-time consistency guarantee mechanism.

[0012] In at least one possible implementation, the downgrading of the current communication mode to another interaction mode includes: if the current communication mode is a mobile communication network, triggering downgrading to a V2X or LoRa communication mode when it is detected that the continuous packet loss rate exceeds a corresponding threshold or the delay time exceeds a corresponding threshold.

[0013] In at least one possible implementation, the downgrading of the current communication mode to another interaction mode includes: if the current communication mode is V2X, triggering downgrading to a LoRa communication mode when it is detected that an RSU node is invalid or the channel utilization rate exceeds a corresponding threshold.

[0014] In at least one possible implementation, the downgrading of the current communication mode to another interaction mode includes: if the current communication mode is LoRa, triggering the start of a vehicle-end visual data interaction mode when it is detected that the LoRa signal strength is lower than a corresponding threshold and lasts for a predetermined length of time.

[0015] In at least one possible implementation, the vehicle-end visual data interaction mode includes:

[0016] Real-time recognition of traffic information by a vehicle-mounted camera using a pre-trained visual perception model, and autonomous path planning and lane-changing decision making in combination with locally stored map data and a vehicle dynamics model;

[0017] Meanwhile, the current communication state is prompted to the driver, and the vehicle control right is handed over to the driver when a signal input by the driver to take over the vehicle is received.

[0018] In at least one possible implementation, the path planning decision level involving vehicle lane-changing of the dynamic allocation includes:

[0019] When the cloud communication delay exceeds a predetermined limit or the edge layer computing resource usage rate reaches a preset interval, an edge layer takeover level is executed: the edge layer takes over the path planning involving lane-changing, and the corresponding path trajectory is executed by the vehicle end.

[0020] In at least one possible implementation, the checking of the trajectory autonomously generated by the vehicle end with the global path includes:

[0021] After the communication quality is restored, the path trajectory autonomously generated by the vehicle end is checked by the cloud;

[0022] If the check result represents that the deviation is within the preset acceptable range, the local path generated by the vehicle end is gradually fused into the latest global path; if the check result represents that an irreversible deviation occurs, global path re-planning is triggered.

[0023] Compared with the prior art, the main design concept of the application is to adopt a hierarchical decision-making vehicle-road cloud cooperative lane changing mode to solve the problem of lane changing interruption caused by unstable communication link and inability to complete the lane changing task; and to solve the contradiction between insufficient computing power and limited vehicle end resources in high-density scenarios by dynamically allocating computing power; further, to combine hierarchical fault fallback and spatiotemporal consistency guarantee mechanism to solve the safety problems such as lateral control shock, too large deviation of vehicle end autonomous decision-making from the global path during communication interruption, and difficulty in rapid correction after recovery. Specifically, the communication link is decomposed into multiple signals, dynamically switched according to real-time signal quality, seamless degradation is achieved, and based on the three-level elastic architecture of cloud end global planning, edge layer regional cooperation and vehicle end local execution, the communication delay and computing power allocation contradiction are avoided by using communication degradation control and computing power resource dynamic allocation strategy. In addition, based on the vehicle-road cloud cooperative architecture, combined with the spatiotemporal consistency guarantee scheme, the vehicle end can still generate a safe lane changing trajectory during communication interruption. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to make the objects, technical solutions and advantages of the present application clearer, the following will further describe the present application with reference to the accompanying drawings, in which:

[0025] Figure 1 A schematic diagram of the vehicle path planning interaction method based on vehicle-road cloud cooperation provided for the embodiments of the present application. DETAILED DESCRIPTION

[0026] The embodiments of the present application will be described in detail below, and examples of the embodiments are shown in the accompanying drawings, in which the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application.

[0027] Based on the analysis of the foregoing background problems, the present application proposes a vehicle-road cloud cooperative vehicle path planning control architecture based on the following, thereby first introducing the architecture scheme.

[0028] The architecture mainly consists of three modules of cloud, edge layer and vehicle end, and the functions of each module are as follows:

[0029] 1. Cloud (global planning and resource scheduling)

[0030] Core functions:

[0031] - High-precision map service: Provides static data such as road topology, slope, and charging station locations;

[0032] -Global route planning module: Generates global routes based on delivery time windows and cargo priorities;

[0033] - Federated learning model aggregation: Receive local model parameters uploaded from the edge layer and update the global lane-changing strategy;

[0034] - Dynamic resource allocation: Adjust the priority of edge layer computing power allocation based on regional traffic load.

[0035] 2. Edge Layer (Regional Collaboration and Real-Time Decision Making)

[0036] Core functions:

[0037] -RSU roadside unit cluster: Deployed at intersections / key road sections, it supports multi-protocol communication (5G / V2X / LoRa), provides lane-level perception data (such as vehicle density, construction area), and performs local perception and communication relay;

[0038] -Local traffic semantic model: Generates a real-time risk field and marks high-risk lane-changing areas;

[0039] - Game theory conflict resolution: predict the intentions of surrounding vehicles and generate conflict-free lane-changing time windows (such as "cut-in gap after 2 seconds").

[0040] - Tiered fault fallback control: Monitor cloud communication status and trigger degradation strategies (such as switching to LoRa to transmit critical data when the network is weak).

[0041] 3. Vehicle-side (Local Execution and Emergency Autonomy)

[0042] Core functions:

[0043] - Multi-source sensor fusion: Fusion of data from lidar, cameras, and load sensors to estimate vehicle status in real time;

[0044] - Lightweight reinforcement learning model: Generates local lane-changing trajectories based on policy parameters distributed from the cloud;

[0045] - Dynamic load adaptation controller: Adjusts lane changing parameters (such as acceleration limit and safety distance) according to real-time load.

[0046] - Visual-assisted communication decoding: Identify roadside LED coded signals and obtain lane status information when the network is down.

[0047] Based on the above architecture description, the interaction flow when communication is normal will be described in detail below:

[0048] (a) Data Uplink Process

[0049] 1) Vehicle-side data acquisition and preprocessing: Multiple types of sensors deployed on the vehicle (such as 128-line LiDAR, high-definition cameras, millimeter-wave radar, and inertial navigation systems) collect real-time information about the vehicle's surrounding environment, including obstacle location, speed, type (pedestrians, vehicles, non-motorized vehicles, etc.), lane line information, vehicle speed, and attitude. The vehicle-side multi-sensor fusion module performs preliminary noise reduction and filtering on the raw data, and transmits it to the edge layer and cloud via the onboard unit (OBU) using 5G or C-V2X communication protocols.

[0050] 2) Edge Layer Semantic Modeling: After receiving data from the vehicle, edge computing nodes use deep learning algorithms (such as BEVFormer and YOLOv8) to perform semantic parsing and modeling. For example, LiDAR point cloud data and camera image data are fused to generate a 3D semantic map of the vehicle's surrounding environment, labeling different target categories, bounding boxes, confidence levels, and other information. Simultaneously, combined with traffic information collected by roadside RSUs (such as traffic light status, traffic events, road construction, etc.), the traffic situation in local areas is analyzed to form a local situation model that includes environmental semantics, traffic rules, and dynamic events. Finally, this model is uploaded to the cloud.

[0051] 3) Cloud-based Global Optimization: The cloud receives local model parameters (such as risk field weights and conflict resolution strategies) uploaded from multiple edge nodes, and updates the global model through gradient difference aggregation algorithms to improve the generalization ability of lane-changing strategies. Spatiotemporal Joint Path Planning: Integrates high-precision maps, real-time traffic flow, and logistics order data to generate globally optimal paths; dynamically adjusts path weights (e.g., reducing energy consumption weights for nighttime deliveries and prioritizing well-lit lanes); dynamically allocates resources: Based on the Traffic Load Index (TLI), additional computing resources are allocated to edge nodes in high-load areas. For example, based on the location, load, and destination of all logistics vehicles, the optimal global path for each vehicle is calculated, and an optimization scheme including lane-changing points, speed suggestions, and time windows is generated.

[0052] (II) Command Downlink Flow

[0053] 1) Cloud-based global path generation and distribution: Based on the above global optimization results, the cloud generates a global path instruction for each vehicle. The instruction includes a detailed sequence of driving route points, key lane-change node information (target lane, lane-change time window, expected speed), and priority parameters. The instruction is encrypted and distributed to the edge computing nodes of the corresponding road segment via C-V2X or 5G network, and the instruction is backed up to the cloud database for subsequent traceability.

[0054] 2) Edge Layer Conflict Resolution: After receiving global path instructions from the cloud, the edge layer combines real-time local traffic information (such as current lane vehicle density, sudden traffic events, and the driving intentions of adjacent vehicles) to perform local conflict detection and resolution on the global path. For example, if dense traffic is detected at lane-changing points planned in the cloud, the edge layer predicts the intersection points of multiple vehicle trajectories within the next 5 seconds and calculates the optimal lane-changing time window based on game theory, adjusting speed recommendations to ensure that the lane-changing process does not conflict with other vehicles. Simultaneously, the edge layer incorporates traffic rule constraints (such as prohibition of changing lanes over solid lines and restrictions during specific time periods), combined with load limits (such as vehicles >10t being prohibited from using emergency lanes) and gradient requirements (such as electric logistics vehicles having a gradient <5%), into the instruction optimization process. The processed instructions include corrected lane-changing timing, speed limits, safety distances, and other local constraint information, and are then sent to the vehicle.

[0055] Vehicle-side trajectory execution: The vehicle-side autonomous driving system receives the lane-changing time window and constraints from the edge layer and parses and verifies the trajectory instructions. The vehicle-side controller, combined with real-time perception data from onboard sensors, determines whether the current environment meets the lane-changing conditions (such as the distance to vehicles behind in the target lane and whether the vehicle's speed is within the allowable range). If the conditions are met, the vehicle's actuators execute the lane-changing operation according to the planned trajectory. During the lane-changing process, the vehicle continuously monitors the surrounding environment. In the event of an emergency (such as another vehicle forcibly changing lanes), it immediately triggers emergency obstacle avoidance or braking strategies to ensure driving safety and feeds back the execution status (actual trajectory, energy consumption, cargo box vibration data) to the edge layer and the cloud. The cloud updates the federated learning model based on the feedback data, optimizes the path planning for the next cycle, and feeds back to the edge layer in real time.

[0056] Based on the foregoing, this invention proposes an embodiment of a vehicle path planning interaction method based on vehicle-road-cloud collaboration. Specifically, as follows: Figure 1 As shown, it includes:

[0057] Step S1: After the autonomous driving mode is activated, if a communication link quality degradation is detected, the communication mode is dynamically switched according to the preset communication priority: the current communication mode is downgraded to another data interaction mode.

[0058] Specifically, when a communication interruption is detected between the vehicle and the edge layer or the cloud, or between the edge layer and the cloud, the communication method is dynamically switched according to the preset communication priority: the current communication method is downgraded to another data interaction method.

[0059] Both the vehicle-side and edge layer are equipped with communication health monitoring (signal strength, packet loss rate, latency). When the vehicle-side does not receive instructions from the edge layer or the cloud, it is determined to be a communication interruption (not just an absolute and complete network outage, but can be understood as a degradation of the communication link quality, including the complete disconnection of the wireless connection). If the edge layer cannot establish a connection with the cloud or the instruction reception times out, a communication interruption alarm will also be triggered.

[0060] When communication link quality deteriorates, the edge layer's tiered fault fallback mechanism can dynamically fall back to the signal interaction method according to a preset communication priority, such as: 5G > V2X > LoRa > visual signals. See the table below for specific implementation examples:

[0061]

[0062] The vehicle-side vision-assisted autonomous driving mode can be further elaborated here: When a wireless communication failure is detected between the vehicle and the edge layer and cloud, the system switches to machine vision-assisted driving mode, temporarily planning a driving path based on visual signals to replace the interaction with the edge layer and cloud in this situation. In actual operation, the onboard camera can use a pre-trained visual perception model to identify traffic information such as lane lines, obstacles, and traffic signs in real time. Combined with locally stored map data and vehicle dynamics models, it can perform autonomous path planning and lane-changing decisions based on the identified roadside LED coded signals or the taillights of the vehicle in front. For example, if the vehicle determines through visual recognition that the target lane is free of obstacles and is a dashed line, it can autonomously perform a lane-changing operation. At the same time, the vehicle can also notify the driver of the current autonomous mode with an audible and visual alarm. If the driver chooses to take over the vehicle, the autonomous driving system will smoothly transfer control to the driver. After communication is restored, the vehicle and the edge layer automatically synchronize data and return to normal collaborative mode.

[0063] Step S2: Based on the degraded communication quality and / or computing resource status, dynamically allocate the path planning decision level involving vehicle lane changing;

[0064] Step S3: Based on the assigned path planning decision level, decide whether to execute the vehicle-side autonomous mode, whereby the vehicle-side autonomous mode refers to the vehicle autonomously planning the path and lane-changing strategy.

[0065] The following table provides examples of preset hierarchical levels:

[0066]

[0067] This section elaborates on the edge layer's dominant layer: Upon detecting a communication interruption with the cloud, the edge layer immediately activates a local emergency strategy, primarily taken over by the edge layer itself. Specifically, it utilizes the latest cached global path information and local real-time perception data to continuously perform local path planning and conflict resolution, ensuring that vehicles within its jurisdiction can still perform safe lane-changing operations based on existing information. For example, the edge layer dynamically adjusts the timing and speed of lane changes based on roadside RSU and onboard sensor data to avoid traffic chaos caused by communication interruptions. Simultaneously, the edge layer records the communication interruption status and local handling solutions in its local log and attempts to re-establish the connection with the cloud.

[0068] Continuing from the previous text, the present invention also includes step S4: when it is determined that the vehicle-side autonomous mode is to be executed, the vehicle-side autonomously generated trajectory is verified with the global path through a spatiotemporal consistency guarantee mechanism;

[0069] This mainly includes the following two aspects:

[0070] First, improve the accuracy of the path trajectory generated autonomously by the vehicle during communication interruptions. Specifically, the vehicle is pre-loaded with lightweight digital twin models of key road sections, including lane topology, gradient, and speed limits. By comparing the autonomously generated local trajectory with the pre-loaded model, progressive correction is triggered when lateral or longitudinal deviations exceed thresholds. Simultaneously, multi-source positioning schemes such as GPS (when available), visual odometry (VIO), and UWB anchor points can be employed, and after Kalman filtering fusion, the accuracy of the basic data for the trajectory generated by the vehicle can be ensured.

[0071] Secondly, after communication is restored, the cloud verifies the path trajectory generated autonomously by the vehicle. Specifically, once communication quality is restored, the vehicle uploads its local trajectory logs to the cloud. The cloud then uses a spatiotemporal cube model to verify the spatiotemporal consistency between the trajectory and the original global path. If the deviation is within acceptable limits (e.g., position deviation <0.5m, timestamp error <2 seconds), it is progressively merged into the latest global path. If irreversible deviations occur (e.g., exceeding the threshold three consecutive times), global replanning is triggered. Based on this concept, it can also be explained that the cloud can periodically distribute "critical paths" (standard trajectory parameters for critical lane change points) to verify the vehicle's local decisions. If the deviation between the real-time trajectory and the critical path exceeds 15%, the system forcibly enters a safe stopping mode, progressively decelerating to an emergency area and activating hazard lights to ensure driving safety in extreme scenarios.

[0072] Those skilled in the art will understand that, following the decision-making hierarchy mentioned above, the safe parking mode can be integrated and set to Level 3. Generally, in addition to exceeding the trajectory deviation limit, it may also be triggered by the failure of vehicle-side sensors, thereby controlling the vehicle to gradually decelerate to a safe area (prioritizing the emergency lane / shoulder) and activating hazard lights, a red warning on the instrument panel, and prompting the driver to take over. Furthermore, fault codes can be sent to edge layer nodes via LoRa communication. For example, the gradual deceleration strategy mentioned here can be implemented by gradually braking to a stop at a deceleration rate of 0.3g.

[0073] In summary, the main design concept of this invention lies in decomposing the communication link into multiple signals and dynamically switching them according to real-time signal quality to achieve seamless degradation. Based on a three-tiered elastic architecture of cloud-based global planning, edge-layer regional collaboration, and vehicle-side local execution, it utilizes communication degradation control and dynamic allocation strategies for computing resources to avoid conflicts arising from communication latency and computing power allocation. Furthermore, based on a vehicle-road-cloud collaborative architecture and combined with a spatiotemporal consistency guarantee scheme, it addresses the safety issue of excessive deviation between vehicle-side autonomous decision-making and the global path, making rapid correction difficult after communication recovery, while ensuring the vehicle can autonomously change lanes during communication interruptions. This invention significantly optimizes the cost-effectiveness of autonomous driving algorithms and effectively improves driving safety, specifically addressing the challenges of unstable communication links combined with limited computing resources.

[0074] In this invention, when directional terms are mentioned, they are relative concepts based on the embodiments. Furthermore, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0075] The above description of the structure, features, and effects of the present invention is based on the embodiments shown in the figures. However, the above are only preferred embodiments of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched by those skilled in the art to form a variety of equivalent solutions without departing from or changing the design concept and technical effects of the present invention. Therefore, the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. A vehicle path planning interaction method based on vehicle-road-cloud collaboration, characterized in that, include: During autonomous driving, if a communication link quality degradation is detected, the communication mode will be dynamically switched according to the preset communication priority: the current communication mode will be downgraded to another interaction mode. Based on the degraded communication quality and / or computing resource status, dynamically allocate the path planning decision level involving vehicle lane changing; Based on the assigned path planning decision level, it is determined whether to execute the vehicle-side autonomous mode, which refers to the vehicle autonomously planning the driving path and lane-changing strategy. If the vehicle-side autonomous mode is determined to be implemented, the trajectory autonomously generated by the vehicle will be verified against the global path through a spatiotemporal consistency guarantee mechanism.

2. The vehicle path planning interaction method based on vehicle-road-cloud collaboration according to claim 1, characterized in that, The step of downgrading the current communication method to other interaction methods includes: if the current communication method is a mobile communication network, when the continuous packet loss rate exceeds the corresponding threshold or the delay time exceeds the corresponding threshold, triggering a downgrade to V2X or LoRa communication method.

3. The vehicle path planning interaction method based on vehicle-road-cloud collaboration according to claim 1, characterized in that, The step of downgrading the current communication method to another interaction method includes: if the current communication method is V2X, when an RSU node failure is detected or the channel utilization exceeds the corresponding threshold, a downgrade to LoRa communication method is triggered.

4. The vehicle path planning interaction method based on vehicle-road-cloud collaboration according to claim 1, characterized in that, The step of downgrading the current communication method to another interaction method includes: if the current communication method is LoRa, when the LoRa signal strength is detected to be lower than the corresponding threshold and lasts for a predetermined duration, the vehicle-side visual data interaction mode is triggered to start.

5. The vehicle path planning interaction method based on vehicle-road-cloud collaboration according to claim 4, characterized in that, The vehicle-side visual data interaction mode includes: Using a pre-trained visual perception model, the vehicle-mounted camera identifies traffic information in real time, and combines it with map data stored locally on the vehicle and vehicle dynamics model to autonomously plan routes and make lane-changing decisions. At the same time, it prompts the driver about the current communication status and, when it receives a signal from the driver to take over the vehicle, transfers control of the vehicle to the driver.

6. The vehicle path planning interaction method based on vehicle-road-cloud collaboration according to claim 1, characterized in that, The dynamic allocation involves path planning decision-making levels for vehicle lane changes, including: When the cloud communication latency exceeds the predetermined limit, or the edge layer computing resource utilization rate reaches the preset range, the edge layer takeover level is executed: the edge layer takes over the path planning involving lane changes, and the vehicle executes the corresponding path trajectory.

7. The vehicle path planning interaction method based on vehicle-road-cloud collaboration according to any one of claims 1 to 6, characterized in that, The step of verifying the vehicle-generated trajectory against the global path includes: After communication quality is restored, the cloud verifies the path trajectory generated autonomously by the vehicle. If the deviation in the verification result is within the preset acceptable range, the local path generated by the vehicle will be gradually merged into the latest global path; if the deviation in the verification result is irreversible, global path replanning will be triggered.

Images